4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2012, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/obdclass/obd_mount_server.c
38 * Server mount routines
40 * Author: Nathan Rutman <nathan@clusterfs.com>
44 #define DEBUG_SUBSYSTEM S_CLASS
45 #define D_MOUNT (D_SUPER | D_CONFIG /* | D_WARNING */)
46 #define PRINT_CMD CDEBUG
47 #define PRINT_MASK (D_SUPER | D_CONFIG)
51 #include <lustre_fsfilt.h>
52 #include <obd_class.h>
53 #include <lustre/lustre_user.h>
54 #include <linux/version.h>
55 #include <lustre_log.h>
56 #include <lustre_disk.h>
57 #include <lustre_param.h>
58 #ifdef HAVE_KERNEL_LOCKED
59 #include <linux/smp_lock.h>
62 /*********** mount lookup *********/
64 DEFINE_MUTEX(lustre_mount_info_lock);
65 static CFS_LIST_HEAD(server_mount_info_list);
67 static struct lustre_mount_info *server_find_mount(const char *name)
70 struct lustre_mount_info *lmi;
73 cfs_list_for_each(tmp, &server_mount_info_list) {
74 lmi = cfs_list_entry(tmp, struct lustre_mount_info,
76 if (strcmp(name, lmi->lmi_name) == 0)
82 /* we must register an obd for a mount before we call the setup routine.
83 *_setup will call lustre_get_mount to get the mnt struct
84 by obd_name, since we can't pass the pointer to setup. */
85 static int server_register_mount(const char *name, struct super_block *sb,
88 struct lustre_mount_info *lmi;
94 OBD_ALLOC(lmi, sizeof(*lmi));
97 OBD_ALLOC(name_cp, strlen(name) + 1);
99 OBD_FREE(lmi, sizeof(*lmi));
102 strcpy(name_cp, name);
104 mutex_lock(&lustre_mount_info_lock);
106 if (server_find_mount(name)) {
107 mutex_unlock(&lustre_mount_info_lock);
108 OBD_FREE(lmi, sizeof(*lmi));
109 OBD_FREE(name_cp, strlen(name) + 1);
110 CERROR("Already registered %s\n", name);
113 lmi->lmi_name = name_cp;
116 cfs_list_add(&lmi->lmi_list_chain, &server_mount_info_list);
118 mutex_unlock(&lustre_mount_info_lock);
120 CDEBUG(D_MOUNT, "reg_mnt %p from %s\n", lmi->lmi_mnt, name);
125 /* when an obd no longer needs a mount */
126 static int server_deregister_mount(const char *name)
128 struct lustre_mount_info *lmi;
131 mutex_lock(&lustre_mount_info_lock);
132 lmi = server_find_mount(name);
134 mutex_unlock(&lustre_mount_info_lock);
135 CERROR("%s not registered\n", name);
139 CDEBUG(D_MOUNT, "dereg_mnt %p from %s\n", lmi->lmi_mnt, name);
141 OBD_FREE(lmi->lmi_name, strlen(lmi->lmi_name) + 1);
142 cfs_list_del(&lmi->lmi_list_chain);
143 OBD_FREE(lmi, sizeof(*lmi));
144 mutex_unlock(&lustre_mount_info_lock);
149 /* obd's look up a registered mount using their obdname. This is just
150 for initial obd setup to find the mount struct. It should not be
151 called every time you want to mntget. */
152 struct lustre_mount_info *server_get_mount(const char *name)
154 struct lustre_mount_info *lmi;
155 struct lustre_sb_info *lsi;
158 mutex_lock(&lustre_mount_info_lock);
159 lmi = server_find_mount(name);
160 mutex_unlock(&lustre_mount_info_lock);
162 CERROR("Can't find mount for %s\n", name);
165 lsi = s2lsi(lmi->lmi_sb);
167 cfs_atomic_inc(&lsi->lsi_mounts);
169 CDEBUG(D_MOUNT, "get_mnt %p from %s, refs=%d\n", lmi->lmi_mnt,
170 name, cfs_atomic_read(&lsi->lsi_mounts));
174 EXPORT_SYMBOL(server_get_mount);
177 * Used by mdt to get mount_info from obdname.
178 * There are no blocking when using the mount_info.
179 * Do not use server_get_mount for this purpose.
181 struct lustre_mount_info *server_get_mount_2(const char *name)
183 struct lustre_mount_info *lmi;
186 mutex_lock(&lustre_mount_info_lock);
187 lmi = server_find_mount(name);
188 mutex_unlock(&lustre_mount_info_lock);
190 CERROR("Can't find mount for %s\n", name);
194 EXPORT_SYMBOL(server_get_mount_2);
196 /* to be called from obd_cleanup methods */
197 int server_put_mount(const char *name, struct vfsmount *mnt)
199 struct lustre_mount_info *lmi;
200 struct lustre_sb_info *lsi;
203 mutex_lock(&lustre_mount_info_lock);
204 lmi = server_find_mount(name);
205 mutex_unlock(&lustre_mount_info_lock);
207 CERROR("Can't find mount for %s\n", name);
210 lsi = s2lsi(lmi->lmi_sb);
212 CDEBUG(D_MOUNT, "put_mnt %p from %s, refs=%d\n",
213 lmi->lmi_mnt, name, cfs_atomic_read(&lsi->lsi_mounts));
215 if (lustre_put_lsi(lmi->lmi_sb))
216 CDEBUG(D_MOUNT, "Last put of mnt %p from %s\n",
219 /* this obd should never need the mount again */
220 server_deregister_mount(name);
224 EXPORT_SYMBOL(server_put_mount);
226 /* Corresponding to server_get_mount_2 */
227 int server_put_mount_2(const char *name, struct vfsmount *mnt)
232 EXPORT_SYMBOL(server_put_mount_2);
234 /* Set up a MGS to serve startup logs */
235 static int server_start_mgs(struct super_block *sb)
237 struct lustre_sb_info *lsi = s2lsi(sb);
238 struct vfsmount *mnt = lsi->lsi_srv_mnt;
239 struct lustre_mount_info *lmi;
243 /* It is impossible to have more than 1 MGS per node, since
244 MGC wouldn't know which to connect to */
245 lmi = server_find_mount(LUSTRE_MGS_OBDNAME);
247 lsi = s2lsi(lmi->lmi_sb);
248 LCONSOLE_ERROR_MSG(0x15d, "The MGS service was already started"
253 CDEBUG(D_CONFIG, "Start MGS service %s\n", LUSTRE_MGS_OBDNAME);
255 rc = server_register_mount(LUSTRE_MGS_OBDNAME, sb, mnt);
258 rc = lustre_start_simple(LUSTRE_MGS_OBDNAME, LUSTRE_MGS_NAME,
259 LUSTRE_MGS_OBDNAME, 0, 0,
260 lsi->lsi_osd_obdname, 0);
261 /* Do NOT call server_deregister_mount() here. This leads to
262 * inability cleanup cleanly and free lsi and other stuff when
263 * mgs calls server_put_mount() in error handling case. -umka */
267 LCONSOLE_ERROR_MSG(0x15e, "Failed to start MGS '%s' (%d). "
268 "Is the 'mgs' module loaded?\n",
269 LUSTRE_MGS_OBDNAME, rc);
273 static int server_stop_mgs(struct super_block *sb)
275 struct obd_device *obd;
279 CDEBUG(D_MOUNT, "Stop MGS service %s\n", LUSTRE_MGS_OBDNAME);
281 /* There better be only one MGS */
282 obd = class_name2obd(LUSTRE_MGS_OBDNAME);
284 CDEBUG(D_CONFIG, "mgs %s not running\n", LUSTRE_MGS_OBDNAME);
288 /* The MGS should always stop when we say so */
290 rc = class_manual_cleanup(obd);
294 /* Since there's only one mgc per node, we have to change it's fs to get
295 access to the right disk. */
296 static int server_mgc_set_fs(struct obd_device *mgc, struct super_block *sb)
298 struct lustre_sb_info *lsi = s2lsi(sb);
302 CDEBUG(D_MOUNT, "Set mgc disk for %s\n", lsi->lsi_lmd->lmd_dev);
304 /* cl_mgc_sem in mgc insures we sleep if the mgc_fs is busy */
305 rc = obd_set_info_async(NULL, mgc->obd_self_export,
306 sizeof(KEY_SET_FS), KEY_SET_FS,
307 sizeof(*sb), sb, NULL);
309 CERROR("can't set_fs %d\n", rc);
314 static int server_mgc_clear_fs(struct obd_device *mgc)
319 CDEBUG(D_MOUNT, "Unassign mgc disk\n");
321 rc = obd_set_info_async(NULL, mgc->obd_self_export,
322 sizeof(KEY_CLEAR_FS), KEY_CLEAR_FS,
327 static int is_mdc_device(const char *devname)
331 ptr = strrchr(devname, '-');
332 if (ptr != NULL && strcmp(ptr, "-mdc") == 0)
338 static inline int tgt_is_mdt0(const char *tgtname)
343 type = server_name2index(tgtname, &idx, NULL);
344 if (type != LDD_F_SV_TYPE_MDT)
350 static inline int is_mdc_for_mdt0(const char *devname)
354 if (!is_mdc_device(devname))
357 ptr = strrchr(devname, '-');
362 if (tgt_is_mdt0(devname)) {
371 * Convert OST/MDT name(fsname-OSTxxxx) to a lwp name
372 * (fsname-MDT0000-lwp-OSTxxxx)
374 int tgt_name2lwpname(const char *svname, char *lwpname)
381 OBD_ALLOC(fsname, MTI_NAME_MAXLEN);
385 rc = server_name2fsname(svname, fsname, &tgt);
387 CERROR("%s: failed to get fsname from svname. %d\n",
392 if (*tgt != '-' && *tgt != ':') {
393 CERROR("%s: invalid svname name!\n", svname);
394 GOTO(cleanup, rc = -EINVAL);
398 if (strncmp(tgt, "OST", 3) != 0 && strncmp(tgt, "MDT", 3) != 0) {
399 CERROR("%s is not an OST or MDT target!\n", svname);
400 GOTO(cleanup, rc = -EINVAL);
402 sprintf(lwpname, "%s-MDT0000-%s-%s", fsname, LUSTRE_LWP_NAME, tgt);
405 OBD_FREE(fsname, MTI_NAME_MAXLEN);
408 EXPORT_SYMBOL(tgt_name2lwpname);
410 static CFS_LIST_HEAD(lwp_register_list);
411 DEFINE_MUTEX(lwp_register_list_lock);
413 int lustre_register_lwp_item(const char *lwpname, struct obd_export **exp,
414 register_lwp_cb cb_func, void *cb_data)
416 struct obd_device *lwp;
417 struct lwp_register_item *lri;
420 LASSERTF(strlen(lwpname) < MTI_NAME_MAXLEN, "lwpname is too long %s\n",
422 LASSERT(exp != NULL && *exp == NULL);
428 mutex_lock(&lwp_register_list_lock);
430 lwp = class_name2obd(lwpname);
431 if (lwp != NULL && lwp->obd_set_up == 1) {
432 struct obd_uuid *uuid;
436 mutex_unlock(&lwp_register_list_lock);
440 memcpy(uuid->uuid, lwpname, strlen(lwpname));
441 *exp = cfs_hash_lookup(lwp->obd_uuid_hash, uuid);
445 memcpy(lri->lri_name, lwpname, strlen(lwpname));
447 lri->lri_cb_func = cb_func;
448 lri->lri_cb_data = cb_data;
449 CFS_INIT_LIST_HEAD(&lri->lri_list);
450 cfs_list_add(&lri->lri_list, &lwp_register_list);
452 if (*exp != NULL && cb_func != NULL)
455 mutex_unlock(&lwp_register_list_lock);
458 EXPORT_SYMBOL(lustre_register_lwp_item);
460 void lustre_deregister_lwp_item(struct obd_export **exp)
462 struct lwp_register_item *lri, *tmp;
464 mutex_lock(&lwp_register_list_lock);
465 cfs_list_for_each_entry_safe(lri, tmp, &lwp_register_list, lri_list) {
466 if (exp == lri->lri_exp) {
468 class_export_put(*exp);
469 cfs_list_del(&lri->lri_list);
474 mutex_unlock(&lwp_register_list_lock);
476 EXPORT_SYMBOL(lustre_deregister_lwp_item);
478 static void lustre_notify_lwp_list(struct obd_export *exp)
480 struct lwp_register_item *lri, *tmp;
481 LASSERT(exp != NULL);
483 mutex_lock(&lwp_register_list_lock);
484 cfs_list_for_each_entry_safe(lri, tmp, &lwp_register_list, lri_list) {
485 if (strcmp(exp->exp_obd->obd_name, lri->lri_name))
487 if (*lri->lri_exp != NULL)
489 *lri->lri_exp = class_export_get(exp);
490 if (lri->lri_cb_func != NULL)
491 lri->lri_cb_func(lri->lri_cb_data);
493 mutex_unlock(&lwp_register_list_lock);
496 static int lustre_lwp_connect(struct obd_device *lwp)
499 struct lu_context session_ctx;
500 struct obd_export *exp;
501 struct obd_uuid *uuid = NULL;
502 struct obd_connect_data *data = NULL;
506 /* log has been fully processed, let clients connect */
507 rc = lu_env_init(&env, lwp->obd_lu_dev->ld_type->ldt_ctx_tags);
511 lu_context_init(&session_ctx, LCT_SESSION);
512 session_ctx.lc_thread = NULL;
513 lu_context_enter(&session_ctx);
514 env.le_ses = &session_ctx;
518 GOTO(out, rc = -ENOMEM);
520 data->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_INDEX;
521 data->ocd_version = LUSTRE_VERSION_CODE;
522 data->ocd_connect_flags |= OBD_CONNECT_MDS_MDS | OBD_CONNECT_FID |
523 OBD_CONNECT_AT | OBD_CONNECT_LRU_RESIZE |
524 OBD_CONNECT_FULL20 | OBD_CONNECT_LVB_TYPE |
525 OBD_CONNECT_LIGHTWEIGHT |
526 OBD_CONNECT_PINGLESS;
529 GOTO(out, rc = -ENOMEM);
531 if (strlen(lwp->obd_name) > sizeof(uuid->uuid)) {
532 CERROR("%s: Too long lwp name %s, max_size is %d\n",
533 lwp->obd_name, lwp->obd_name, (int)sizeof(uuid->uuid));
534 GOTO(out, rc = -EINVAL);
537 /* Use lwp name as the uuid, so we find the export by lwp name later */
538 memcpy(uuid->uuid, lwp->obd_name, strlen(lwp->obd_name));
539 rc = obd_connect(&env, &exp, lwp, uuid, data, NULL);
541 CERROR("%s: connect failed: rc = %d\n", lwp->obd_name, rc);
543 lustre_notify_lwp_list(exp);
552 lu_context_exit(&session_ctx);
553 lu_context_fini(&session_ctx);
559 * lwp is used by slaves (Non-MDT0 targets) to manage the connection
562 static int lustre_lwp_setup(struct lustre_cfg *lcfg, struct lustre_sb_info *lsi)
564 struct obd_connect_data *data = NULL;
565 struct obd_device *obd;
566 char *lwpname = NULL;
567 char *lwpuuid = NULL;
571 rc = class_add_uuid(lustre_cfg_string(lcfg, 1),
574 CERROR("%s: Can't add uuid: rc =%d\n", lsi->lsi_svname, rc);
578 OBD_ALLOC(lwpname, MTI_NAME_MAXLEN);
580 GOTO(out, rc = -ENOMEM);
582 rc = tgt_name2lwpname(lsi->lsi_svname, lwpname);
584 CERROR("%s: failed to generate lwp name. %d\n",
585 lsi->lsi_svname, rc);
589 OBD_ALLOC(lwpuuid, MTI_NAME_MAXLEN);
591 GOTO(out, rc = -ENOMEM);
593 sprintf(lwpuuid, "%s_UUID", lwpname);
594 rc = lustre_start_simple(lwpname, LUSTRE_LWP_NAME,
595 lwpuuid, lustre_cfg_string(lcfg, 1),
598 CERROR("%s: setup up failed: rc %d\n", lwpname, rc);
602 obd = class_name2obd(lwpname);
603 LASSERT(obd != NULL);
605 rc = lustre_lwp_connect(obd);
607 CERROR("%s: connect failed: rc = %d\n", lwpname, rc);
612 OBD_FREE(lwpname, MTI_NAME_MAXLEN);
614 OBD_FREE(lwpuuid, MTI_NAME_MAXLEN);
619 /* the caller is responsible for memory free */
620 static struct obd_device *lustre_find_lwp(struct lustre_sb_info *lsi,
621 char **lwpname, char **logname)
623 struct obd_device *lwp;
627 LASSERT(lwpname != NULL);
628 LASSERT(IS_OST(lsi) || IS_MDT(lsi));
630 OBD_ALLOC(*lwpname, MTI_NAME_MAXLEN);
631 if (*lwpname == NULL)
632 RETURN(ERR_PTR(-ENOMEM));
634 if (logname != NULL) {
635 OBD_ALLOC(*logname, MTI_NAME_MAXLEN);
636 if (*logname == NULL)
637 GOTO(out, rc = -ENOMEM);
638 rc = server_name2fsname(lsi->lsi_svname, *lwpname, NULL);
640 CERROR("%s: failed to get fsname from svname. %d\n",
641 lsi->lsi_svname, rc);
642 GOTO(out, rc = -EINVAL);
644 sprintf(*logname, "%s-client", *lwpname);
647 rc = tgt_name2lwpname(lsi->lsi_svname, *lwpname);
649 CERROR("%s: failed to generate lwp name. %d\n",
650 lsi->lsi_svname, rc);
651 GOTO(out, rc = -EINVAL);
654 lwp = class_name2obd(*lwpname);
658 if (*lwpname != NULL) {
659 OBD_FREE(*lwpname, MTI_NAME_MAXLEN);
662 if (logname != NULL && *logname != NULL) {
663 OBD_FREE(*logname, MTI_NAME_MAXLEN);
669 RETURN(lwp != NULL ? lwp : ERR_PTR(-ENOENT));
672 static int lustre_lwp_add_conn(struct lustre_cfg *cfg,
673 struct lustre_sb_info *lsi)
675 struct lustre_cfg_bufs *bufs = NULL;
676 struct lustre_cfg *lcfg = NULL;
677 char *lwpname = NULL;
678 struct obd_device *lwp;
682 lwp = lustre_find_lwp(lsi, &lwpname, NULL);
684 CERROR("%s: can't find lwp device.\n", lsi->lsi_svname);
685 GOTO(out, rc = PTR_ERR(lwp));
687 LASSERT(lwpname != NULL);
691 GOTO(out, rc = -ENOMEM);
693 lustre_cfg_bufs_reset(bufs, lwpname);
694 lustre_cfg_bufs_set_string(bufs, 1,
695 lustre_cfg_string(cfg, 1));
697 lcfg = lustre_cfg_new(LCFG_ADD_CONN, bufs);
699 rc = class_add_conn(lwp, lcfg);
701 CERROR("%s: can't add conn: rc = %d\n", lwpname, rc);
707 lustre_cfg_free(lcfg);
709 OBD_FREE(lwpname, MTI_NAME_MAXLEN);
714 * Retrieve MDT nids from the client log, then start the lwp device.
715 * there are only two scenarios which would include mdt nid.
717 * marker 5 (flags=0x01, v2.1.54.0) lustre-MDT0000 'add mdc' xxx-
718 * add_uuid nid=192.168.122.162@tcp(0x20000c0a87aa2) 0: 1:192.168.122.162@tcp
719 * attach 0:lustre-MDT0000-mdc 1:mdc 2:lustre-clilmv_UUID
720 * setup 0:lustre-MDT0000-mdc 1:lustre-MDT0000_UUID 2:192.168.122.162@tcp
721 * add_uuid nid=192.168.172.1@tcp(0x20000c0a8ac01) 0: 1:192.168.172.1@tcp
722 * add_conn 0:lustre-MDT0000-mdc 1:192.168.172.1@tcp
723 * modify_mdc_tgts add 0:lustre-clilmv 1:lustre-MDT0000_UUID xxxx
724 * marker 5 (flags=0x02, v2.1.54.0) lustre-MDT0000 'add mdc' xxxx-
726 * marker 7 (flags=0x01, v2.1.54.0) lustre-MDT0000 'add failnid' xxxx-
727 * add_uuid nid=192.168.122.2@tcp(0x20000c0a87a02) 0: 1:192.168.122.2@tcp
728 * add_conn 0:lustre-MDT0000-mdc 1:192.168.122.2@tcp
729 * marker 7 (flags=0x02, v2.1.54.0) lustre-MDT0000 'add failnid' xxxx-
731 static int client_lwp_config_process(const struct lu_env *env,
732 struct llog_handle *handle,
733 struct llog_rec_hdr *rec, void *data)
735 struct config_llog_instance *clli = data;
736 int cfg_len = rec->lrh_len;
737 char *cfg_buf = (char *) (rec + 1);
738 struct lustre_cfg *lcfg = NULL;
739 struct lustre_sb_info *lsi;
740 int rc = 0, swab = 0;
743 if (rec->lrh_type != OBD_CFG_REC) {
744 CERROR("Unknown llog record type %#x encountered\n",
749 LASSERT(clli->cfg_sb != NULL);
750 lsi = s2lsi(clli->cfg_sb);
752 lcfg = (struct lustre_cfg *)cfg_buf;
753 if (lcfg->lcfg_version == __swab32(LUSTRE_CFG_VERSION)) {
754 lustre_swab_lustre_cfg(lcfg);
758 rc = lustre_cfg_sanity_check(cfg_buf, cfg_len);
762 switch (lcfg->lcfg_command) {
764 struct cfg_marker *marker = lustre_cfg_buf(lcfg, 1);
766 lustre_swab_cfg_marker(marker, swab,
767 LUSTRE_CFG_BUFLEN(lcfg, 1));
768 if (marker->cm_flags & CM_SKIP ||
769 marker->cm_flags & CM_EXCLUDE)
772 if (!tgt_is_mdt0(marker->cm_tgtname))
775 if (!strncmp(marker->cm_comment, "add mdc", 7) ||
776 !strncmp(marker->cm_comment, "add failnid", 11)) {
777 if (marker->cm_flags & CM_START) {
778 clli->cfg_flags = CFG_F_MARKER;
779 /* This hack is to differentiate the
780 * ADD_UUID is come from "add mdc" record
781 * or from "add failnid" record. */
782 if (!strncmp(marker->cm_comment,
784 clli->cfg_flags |= CFG_F_SKIP;
785 } else if (marker->cm_flags & CM_END) {
791 case LCFG_ADD_UUID: {
792 if (clli->cfg_flags == CFG_F_MARKER) {
793 rc = lustre_lwp_setup(lcfg, lsi);
794 /* XXX: process only the first nid as
795 * we don't need another instance of lwp */
796 clli->cfg_flags |= CFG_F_SKIP;
797 } else if (clli->cfg_flags == (CFG_F_MARKER | CFG_F_SKIP)) {
798 rc = class_add_uuid(lustre_cfg_string(lcfg, 1),
801 CERROR("%s: Fail to add uuid, rc:%d\n",
802 lsi->lsi_svname, rc);
806 case LCFG_ADD_CONN: {
807 if (is_mdc_for_mdt0(lustre_cfg_string(lcfg, 0)))
808 rc = lustre_lwp_add_conn(lcfg, lsi);
818 static int lustre_disconnect_lwp(struct super_block *sb)
820 struct lustre_sb_info *lsi = s2lsi(sb);
821 struct obd_device *lwp;
822 char *lwpname = NULL;
823 char *logname = NULL;
824 struct lustre_cfg *lcfg = NULL;
825 struct lustre_cfg_bufs *bufs = NULL;
826 struct config_llog_instance *cfg = NULL;
830 lwp = lustre_find_lwp(lsi, &lwpname, &logname);
831 if (IS_ERR(lwp) && PTR_ERR(lwp) != -ENOENT)
832 GOTO(out, rc = PTR_ERR(lwp));
834 LASSERT(lwpname != NULL);
835 LASSERT(logname != NULL);
839 GOTO(out, rc = -ENOMEM);
842 cfg->cfg_instance = sb;
843 rc = lustre_end_log(sb, logname, cfg);
845 CERROR("%s: Can't end config log %s.\n", lwpname, logname);
849 if (PTR_ERR(lwp) == -ENOENT) {
850 CDEBUG(D_CONFIG, "%s: lwp device wasn't started.\n",
857 GOTO(out, rc = -ENOMEM);
859 lustre_cfg_bufs_reset(bufs, lwp->obd_name);
860 lustre_cfg_bufs_set_string(bufs, 1, NULL);
861 lcfg = lustre_cfg_new(LCFG_CLEANUP, bufs);
863 GOTO(out, rc = -ENOMEM);
865 /* Disconnect import first. NULL is passed for the '@env', since
866 * it will not be used. */
867 rc = lwp->obd_lu_dev->ld_ops->ldo_process_config(NULL, lwp->obd_lu_dev,
871 lustre_cfg_free(lcfg);
877 OBD_FREE(lwpname, MTI_NAME_MAXLEN);
879 OBD_FREE(logname, MTI_NAME_MAXLEN);
884 * Stop the lwp for an OST/MDT target.
886 static int lustre_stop_lwp(struct super_block *sb)
888 struct lustre_sb_info *lsi = s2lsi(sb);
889 struct obd_device *lwp = NULL;
890 char *lwpname = NULL;
894 lwp = lustre_find_lwp(lsi, &lwpname, NULL);
896 CDEBUG(PTR_ERR(lwp) == -ENOENT ? D_CONFIG : D_ERROR,
897 "%s: lwp wasn't started.\n", lsi->lsi_svname);
902 rc = class_manual_cleanup(lwp);
906 OBD_FREE(lwpname, MTI_NAME_MAXLEN);
911 * Start the lwp(fsname-MDT0000-lwp-OSTxxxx) for an OST or MDT target,
912 * which would be used to establish connection from OST to MDT0.
914 static int lustre_start_lwp(struct super_block *sb)
916 struct lustre_sb_info *lsi = s2lsi(sb);
917 struct config_llog_instance *cfg = NULL;
918 struct obd_device *lwp;
919 char *lwpname = NULL;
920 char *logname = NULL;
924 lwp = lustre_find_lwp(lsi, &lwpname, &logname);
926 /* the lwp device already stared */
927 if (lwp && !IS_ERR(lwp))
930 if (PTR_ERR(lwp) != -ENOENT)
931 GOTO(out, rc = PTR_ERR(lwp));
933 LASSERT(lwpname != NULL);
934 LASSERT(logname != NULL);
938 GOTO(out, rc = -ENOMEM);
940 cfg->cfg_callback = client_lwp_config_process;
941 cfg->cfg_instance = sb;
943 rc = lustre_process_log(sb, logname, cfg);
946 OBD_FREE(lwpname, MTI_NAME_MAXLEN);
948 OBD_FREE(logname, MTI_NAME_MAXLEN);
954 DEFINE_MUTEX(server_start_lock);
956 /* Stop MDS/OSS if nobody is using them */
957 static int server_stop_servers(int lsiflags)
959 struct obd_device *obd = NULL;
960 struct obd_type *type = NULL;
964 mutex_lock(&server_start_lock);
966 /* Either an MDT or an OST or neither */
967 /* if this was an MDT, and there are no more MDT's, clean up the MDS */
968 if (lsiflags & LDD_F_SV_TYPE_MDT) {
969 obd = class_name2obd(LUSTRE_MDS_OBDNAME);
971 type = class_search_type(LUSTRE_MDT_NAME);
974 /* if this was an OST, and there are no more OST's, clean up the OSS */
975 if (lsiflags & LDD_F_SV_TYPE_OST) {
976 obd = class_name2obd(LUSTRE_OSS_OBDNAME);
978 type = class_search_type(LUSTRE_OST_NAME);
981 if (obd != NULL && (type == NULL || type->typ_refcnt == 0)) {
985 /* obd_fail doesn't mean much on a server obd */
986 err = class_manual_cleanup(obd);
991 mutex_unlock(&server_start_lock);
996 int server_mti_print(const char *title, struct mgs_target_info *mti)
998 PRINT_CMD(PRINT_MASK, "mti %s\n", title);
999 PRINT_CMD(PRINT_MASK, "server: %s\n", mti->mti_svname);
1000 PRINT_CMD(PRINT_MASK, "fs: %s\n", mti->mti_fsname);
1001 PRINT_CMD(PRINT_MASK, "uuid: %s\n", mti->mti_uuid);
1002 PRINT_CMD(PRINT_MASK, "ver: %d flags: %#x\n",
1003 mti->mti_config_ver, mti->mti_flags);
1006 EXPORT_SYMBOL(server_mti_print);
1008 /* Generate data for registration */
1009 static int server_lsi2mti(struct lustre_sb_info *lsi,
1010 struct mgs_target_info *mti)
1012 lnet_process_id_t id;
1017 if (!IS_SERVER(lsi))
1020 if (strlcpy(mti->mti_svname, lsi->lsi_svname, sizeof(mti->mti_svname))
1021 >= sizeof(mti->mti_svname))
1024 mti->mti_nid_count = 0;
1025 while (LNetGetId(i++, &id) != -ENOENT) {
1026 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
1029 /* server use --servicenode param, only allow specified
1030 * nids be registered */
1031 if ((lsi->lsi_lmd->lmd_flags & LMD_FLG_NO_PRIMNODE) != 0 &&
1032 class_match_nid(lsi->lsi_lmd->lmd_params,
1033 PARAM_FAILNODE, id.nid) < 1)
1036 /* match specified network */
1037 if (!class_match_net(lsi->lsi_lmd->lmd_params,
1038 PARAM_NETWORK, LNET_NIDNET(id.nid)))
1041 mti->mti_nids[mti->mti_nid_count] = id.nid;
1042 mti->mti_nid_count++;
1043 if (mti->mti_nid_count >= MTI_NIDS_MAX) {
1044 CWARN("Only using first %d nids for %s\n",
1045 mti->mti_nid_count, mti->mti_svname);
1050 mti->mti_lustre_ver = LUSTRE_VERSION_CODE;
1051 mti->mti_config_ver = 0;
1053 rc = server_name2fsname(lsi->lsi_svname, mti->mti_fsname, NULL);
1057 rc = server_name2index(lsi->lsi_svname, &mti->mti_stripe_index, NULL);
1060 /* Orion requires index to be set */
1061 LASSERT(!(rc & LDD_F_NEED_INDEX));
1062 /* keep only LDD flags */
1063 mti->mti_flags = lsi->lsi_flags & LDD_F_MASK;
1064 if (mti->mti_flags & (LDD_F_WRITECONF | LDD_F_VIRGIN))
1065 mti->mti_flags |= LDD_F_UPDATE;
1066 cplen = strlcpy(mti->mti_params, lsi->lsi_lmd->lmd_params,
1067 sizeof(mti->mti_params));
1068 if (cplen >= sizeof(mti->mti_params))
1073 /* Register an old or new target with the MGS. If needed MGS will construct
1074 startup logs and assign index */
1075 static int server_register_target(struct lustre_sb_info *lsi)
1077 struct obd_device *mgc = lsi->lsi_mgc;
1078 struct mgs_target_info *mti = NULL;
1085 if (!IS_SERVER(lsi))
1092 rc = server_lsi2mti(lsi, mti);
1096 CDEBUG(D_MOUNT, "Registration %s, fs=%s, %s, index=%04x, flags=%#x\n",
1097 mti->mti_svname, mti->mti_fsname,
1098 libcfs_nid2str(mti->mti_nids[0]), mti->mti_stripe_index,
1101 /* if write_conf is true, the registration must succeed */
1102 writeconf = !!(lsi->lsi_flags & (LDD_F_NEED_INDEX | LDD_F_UPDATE));
1103 mti->mti_flags |= LDD_F_OPC_REG;
1105 /* Register the target */
1106 /* FIXME use mgc_process_config instead */
1107 rc = obd_set_info_async(NULL, mgc->u.cli.cl_mgc_mgsexp,
1108 sizeof(KEY_REGISTER_TARGET),
1109 KEY_REGISTER_TARGET,
1110 sizeof(*mti), mti, NULL);
1112 if (mti->mti_flags & LDD_F_ERROR) {
1113 LCONSOLE_ERROR_MSG(0x160,
1114 "%s: the MGS refuses to allow this server "
1115 "to start: rc = %d. Please see messages on "
1116 "the MGS.\n", lsi->lsi_svname, rc);
1117 } else if (writeconf) {
1118 LCONSOLE_ERROR_MSG(0x15f,
1119 "%s: cannot register this server with the MGS: "
1120 "rc = %d. Is the MGS running?\n",
1121 lsi->lsi_svname, rc);
1123 CERROR("%s: error registering with the MGS: rc = %d "
1124 "(not fatal)\n", lsi->lsi_svname, rc);
1125 /* reset the error code for non-fatal error. */
1138 * Notify the MGS that this target is ready.
1139 * Used by IR - if the MGS receives this message, it will notify clients.
1141 static int server_notify_target(struct super_block *sb, struct obd_device *obd)
1143 struct lustre_sb_info *lsi = s2lsi(sb);
1144 struct obd_device *mgc = lsi->lsi_mgc;
1145 struct mgs_target_info *mti = NULL;
1151 if (!(IS_SERVER(lsi)))
1157 rc = server_lsi2mti(lsi, mti);
1161 mti->mti_instance = obd->u.obt.obt_instance;
1162 mti->mti_flags |= LDD_F_OPC_READY;
1164 /* FIXME use mgc_process_config instead */
1165 rc = obd_set_info_async(NULL, mgc->u.cli.cl_mgc_mgsexp,
1166 sizeof(KEY_REGISTER_TARGET),
1167 KEY_REGISTER_TARGET,
1168 sizeof(*mti), mti, NULL);
1170 /* Imperative recovery: if the mgs informs us to use IR? */
1171 if (!rc && !(mti->mti_flags & LDD_F_ERROR) &&
1172 (mti->mti_flags & LDD_F_IR_CAPABLE))
1173 lsi->lsi_flags |= LDD_F_IR_CAPABLE;
1182 /** Start server targets: MDTs and OSTs
1184 static int server_start_targets(struct super_block *sb, struct vfsmount *mnt)
1186 struct obd_device *obd;
1187 struct lustre_sb_info *lsi = s2lsi(sb);
1188 struct config_llog_instance cfg;
1190 struct lu_device *dev;
1194 CDEBUG(D_MOUNT, "starting target %s\n", lsi->lsi_svname);
1197 /* make sure the MDS is started */
1198 mutex_lock(&server_start_lock);
1199 obd = class_name2obd(LUSTRE_MDS_OBDNAME);
1201 rc = lustre_start_simple(LUSTRE_MDS_OBDNAME,
1203 LUSTRE_MDS_OBDNAME"_uuid",
1206 mutex_unlock(&server_start_lock);
1207 CERROR("failed to start MDS: %d\n", rc);
1211 mutex_unlock(&server_start_lock);
1214 /* If we're an OST, make sure the global OSS is running */
1216 /* make sure OSS is started */
1217 mutex_lock(&server_start_lock);
1218 obd = class_name2obd(LUSTRE_OSS_OBDNAME);
1220 rc = lustre_start_simple(LUSTRE_OSS_OBDNAME,
1222 LUSTRE_OSS_OBDNAME"_uuid",
1225 mutex_unlock(&server_start_lock);
1226 CERROR("failed to start OSS: %d\n", rc);
1230 mutex_unlock(&server_start_lock);
1233 /* Set the mgc fs to our server disk. This allows the MGC to
1234 * read and write configs locally, in case it can't talk to the MGS. */
1235 if (lsi->lsi_srv_mnt) {
1236 rc = server_mgc_set_fs(lsi->lsi_mgc, sb);
1238 GOTO(out_stop_service, rc);
1241 /* Register with MGS */
1242 rc = server_register_target(lsi);
1246 /* Let the target look up the mount using the target's name
1247 (we can't pass the sb or mnt through class_process_config.) */
1248 rc = server_register_mount(lsi->lsi_svname, sb, mnt);
1252 /* Start targets using the llog named for the target */
1253 memset(&cfg, 0, sizeof(cfg));
1254 cfg.cfg_callback = class_config_llog_handler;
1255 rc = lustre_process_log(sb, lsi->lsi_svname, &cfg);
1257 CERROR("failed to start server %s: %d\n",
1258 lsi->lsi_svname, rc);
1259 /* Do NOT call server_deregister_mount() here. This makes it
1260 * impossible to find mount later in cleanup time and leaves
1261 * @lsi and othder stuff leaked. -umka */
1265 obd = class_name2obd(lsi->lsi_svname);
1267 CERROR("no server named %s was started\n", lsi->lsi_svname);
1268 GOTO(out_mgc, rc = -ENXIO);
1271 if (IS_OST(lsi) || IS_MDT(lsi)) {
1272 rc = lustre_start_lwp(sb);
1274 CERROR("%s: failed to start LWP: %d\n",
1275 lsi->lsi_svname, rc);
1280 server_notify_target(sb, obd);
1282 /* calculate recovery timeout, do it after lustre_process_log */
1283 server_calc_timeout(lsi, obd);
1285 /* log has been fully processed */
1286 obd_notify(obd, NULL, OBD_NOTIFY_CONFIG, (void *)CONFIG_LOG);
1288 /* log has been fully processed, let clients connect */
1289 dev = obd->obd_lu_dev;
1290 if (dev && dev->ld_ops->ldo_prepare) {
1291 rc = lu_env_init(&env, dev->ld_type->ldt_ctx_tags);
1293 struct lu_context session_ctx;
1295 lu_context_init(&session_ctx, LCT_SESSION);
1296 session_ctx.lc_thread = NULL;
1297 lu_context_enter(&session_ctx);
1298 env.le_ses = &session_ctx;
1300 rc = dev->ld_ops->ldo_prepare(&env, NULL, dev);
1303 lu_context_exit(&session_ctx);
1304 lu_context_fini(&session_ctx);
1308 /* abort recovery only on the complete stack:
1309 * many devices can be involved */
1310 if ((lsi->lsi_lmd->lmd_flags & LMD_FLG_ABORT_RECOV) &&
1311 (OBP(obd, iocontrol))) {
1312 obd_iocontrol(OBD_IOC_ABORT_RECOVERY, obd->obd_self_export, 0,
1317 /* Release the mgc fs for others to use */
1318 if (lsi->lsi_srv_mnt)
1319 server_mgc_clear_fs(lsi->lsi_mgc);
1323 server_stop_servers(lsi->lsi_flags);
1328 static int lsi_prepare(struct lustre_sb_info *lsi)
1335 LASSERT(lsi->lsi_lmd);
1337 /* The server name is given as a mount line option */
1338 if (lsi->lsi_lmd->lmd_profile == NULL) {
1339 LCONSOLE_ERROR("Can't determine server name\n");
1343 if (strlen(lsi->lsi_lmd->lmd_profile) >= sizeof(lsi->lsi_svname))
1344 RETURN(-ENAMETOOLONG);
1346 strcpy(lsi->lsi_svname, lsi->lsi_lmd->lmd_profile);
1348 /* Determine osd type */
1349 if (lsi->lsi_lmd->lmd_osd_type != NULL) {
1350 if (strlen(lsi->lsi_lmd->lmd_osd_type) >=
1351 sizeof(lsi->lsi_osd_type))
1352 RETURN(-ENAMETOOLONG);
1354 strcpy(lsi->lsi_osd_type, lsi->lsi_lmd->lmd_osd_type);
1356 strcpy(lsi->lsi_osd_type, LUSTRE_OSD_LDISKFS_NAME);
1359 /* XXX: a temp. solution for components using fsfilt
1360 * to be removed in one of the subsequent patches */
1361 if (!strcmp(lsi->lsi_lmd->lmd_osd_type, "osd-ldiskfs"))
1362 strcpy(lsi->lsi_fstype, "ldiskfs");
1364 strcpy(lsi->lsi_fstype, lsi->lsi_lmd->lmd_osd_type);
1366 /* Determine server type */
1367 rc = server_name2index(lsi->lsi_svname, &index, NULL);
1369 if (lsi->lsi_lmd->lmd_flags & LMD_FLG_MGS) {
1370 /* Assume we're a bare MGS */
1372 lsi->lsi_lmd->lmd_flags |= LMD_FLG_NOSVC;
1374 LCONSOLE_ERROR("Can't determine server type of '%s'\n",
1379 lsi->lsi_flags |= rc;
1381 /* Add mount line flags that used to be in ldd:
1382 * writeconf, mgs, anything else?
1384 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_WRITECONF) ?
1385 LDD_F_WRITECONF : 0;
1386 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_VIRGIN) ?
1388 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_MGS) ?
1389 LDD_F_SV_TYPE_MGS : 0;
1390 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_NO_PRIMNODE) ?
1391 LDD_F_NO_PRIMNODE : 0;
1396 /*************** server mount ******************/
1398 /** Start the shutdown of servers at umount.
1400 static void server_put_super(struct super_block *sb)
1402 struct lustre_sb_info *lsi = s2lsi(sb);
1403 struct obd_device *obd;
1404 char *tmpname, *extraname = NULL;
1406 int lsiflags = lsi->lsi_flags;
1409 LASSERT(IS_SERVER(lsi));
1411 tmpname_sz = strlen(lsi->lsi_svname) + 1;
1412 OBD_ALLOC(tmpname, tmpname_sz);
1413 memcpy(tmpname, lsi->lsi_svname, tmpname_sz);
1414 CDEBUG(D_MOUNT, "server put_super %s\n", tmpname);
1415 if (IS_MDT(lsi) && (lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC))
1416 snprintf(tmpname, tmpname_sz, "MGS");
1418 /* disconnect the lwp first to drain off the inflight request */
1419 if (IS_OST(lsi) || IS_MDT(lsi)) {
1422 rc = lustre_disconnect_lwp(sb);
1423 if (rc && rc != ETIMEDOUT)
1424 CERROR("%s: failed to disconnect lwp. (rc=%d)\n",
1428 /* Stop the target */
1429 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC) &&
1430 (IS_MDT(lsi) || IS_OST(lsi))) {
1431 struct lustre_profile *lprof = NULL;
1433 /* tell the mgc to drop the config log */
1434 lustre_end_log(sb, lsi->lsi_svname, NULL);
1436 /* COMPAT_146 - profile may get deleted in mgc_cleanup.
1437 If there are any setup/cleanup errors, save the lov
1438 name for safety cleanup later. */
1439 lprof = class_get_profile(lsi->lsi_svname);
1440 if (lprof && lprof->lp_dt) {
1441 OBD_ALLOC(extraname, strlen(lprof->lp_dt) + 1);
1442 strcpy(extraname, lprof->lp_dt);
1445 obd = class_name2obd(lsi->lsi_svname);
1447 CDEBUG(D_MOUNT, "stopping %s\n", obd->obd_name);
1448 if (lsiflags & LSI_UMOUNT_FAILOVER)
1450 /* We can't seem to give an error return code
1451 * to .put_super, so we better make sure we clean up! */
1453 class_manual_cleanup(obd);
1455 CERROR("no obd %s\n", lsi->lsi_svname);
1456 server_deregister_mount(lsi->lsi_svname);
1460 /* If they wanted the mgs to stop separately from the mdt, they
1461 should have put it on a different device. */
1463 /* if MDS start with --nomgs, don't stop MGS then */
1464 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOMGS))
1465 server_stop_mgs(sb);
1468 if (IS_OST(lsi) || IS_MDT(lsi)) {
1469 if (lustre_stop_lwp(sb) < 0)
1470 CERROR("%s: failed to stop lwp!\n", tmpname);
1473 /* Clean the mgc and sb */
1474 lustre_common_put_super(sb);
1476 /* wait till all in-progress cleanups are done
1477 * specifically we're interested in ofd cleanup
1479 obd_zombie_barrier();
1481 /* Stop the servers (MDS, OSS) if no longer needed. We must wait
1482 until the target is really gone so that our type refcount check
1484 server_stop_servers(lsiflags);
1486 /* In case of startup or cleanup err, stop related obds */
1488 obd = class_name2obd(extraname);
1490 CWARN("Cleaning orphaned obd %s\n", extraname);
1492 class_manual_cleanup(obd);
1494 OBD_FREE(extraname, strlen(extraname) + 1);
1497 LCONSOLE_WARN("server umount %s complete\n", tmpname);
1498 OBD_FREE(tmpname, tmpname_sz);
1502 /** Called only for 'umount -f'
1504 static void server_umount_begin(struct super_block *sb)
1506 struct lustre_sb_info *lsi = s2lsi(sb);
1509 CDEBUG(D_MOUNT, "umount -f\n");
1510 /* umount = failover
1512 no third way to do non-force, non-failover */
1513 lsi->lsi_flags &= ~LSI_UMOUNT_FAILOVER;
1517 static int server_statfs(struct dentry *dentry, struct kstatfs *buf)
1519 struct super_block *sb = dentry->d_sb;
1520 struct lustre_sb_info *lsi = s2lsi(sb);
1521 struct obd_statfs statfs;
1525 if (lsi->lsi_dt_dev) {
1526 rc = dt_statfs(NULL, lsi->lsi_dt_dev, &statfs);
1528 statfs_unpack(buf, &statfs);
1529 buf->f_type = sb->s_magic;
1535 buf->f_type = sb->s_magic;
1536 buf->f_bsize = sb->s_blocksize;
1542 buf->f_namelen = NAME_MAX;
1546 /** The operations we support directly on the superblock:
1547 * mount, umount, and df.
1549 static struct super_operations server_ops = {
1550 .put_super = server_put_super,
1551 .umount_begin = server_umount_begin, /* umount -f */
1552 .statfs = server_statfs,
1555 #define log2(n) ffz(~(n))
1556 #define LUSTRE_SUPER_MAGIC 0x0BD00BD1
1558 static int server_fill_super_common(struct super_block *sb)
1560 struct inode *root = 0;
1563 CDEBUG(D_MOUNT, "Server sb, dev=%d\n", (int)sb->s_dev);
1565 sb->s_blocksize = 4096;
1566 sb->s_blocksize_bits = log2(sb->s_blocksize);
1567 sb->s_magic = LUSTRE_SUPER_MAGIC;
1568 sb->s_maxbytes = 0; /* we don't allow file IO on server mountpoints */
1569 sb->s_flags |= MS_RDONLY;
1570 sb->s_op = &server_ops;
1572 root = new_inode(sb);
1574 CERROR("Can't make root inode\n");
1578 /* returns -EIO for every operation */
1579 /* make_bad_inode(root); -- badness - can't umount */
1580 /* apparently we need to be a directory for the mount to finish */
1581 root->i_mode = S_IFDIR;
1583 sb->s_root = d_make_root(root);
1585 CERROR("%s: can't make root dentry\n", sb->s_id);
1592 static int osd_start(struct lustre_sb_info *lsi, unsigned long mflags)
1594 struct lustre_mount_data *lmd = lsi->lsi_lmd;
1595 struct obd_device *obd;
1596 struct dt_device_param p;
1602 "Attempting to start %s, type=%s, lsifl=%x, mountfl=%lx\n",
1603 lsi->lsi_svname, lsi->lsi_osd_type, lsi->lsi_flags, mflags);
1605 sprintf(lsi->lsi_osd_obdname, "%s-osd", lsi->lsi_svname);
1606 strcpy(lsi->lsi_osd_uuid, lsi->lsi_osd_obdname);
1607 strcat(lsi->lsi_osd_uuid, "_UUID");
1608 sprintf(flagstr, "%lu:%lu", mflags, (unsigned long) lmd->lmd_flags);
1610 obd = class_name2obd(lsi->lsi_osd_obdname);
1612 rc = lustre_start_simple(lsi->lsi_osd_obdname,
1614 lsi->lsi_osd_uuid, lmd->lmd_dev,
1615 flagstr, lsi->lsi_lmd->lmd_opts,
1619 obd = class_name2obd(lsi->lsi_osd_obdname);
1623 rc = obd_connect(NULL, &lsi->lsi_osd_exp,
1624 obd, &obd->obd_uuid, NULL, NULL);
1627 class_manual_cleanup(obd);
1628 lsi->lsi_dt_dev = NULL;
1631 /* XXX: to keep support old components relying on lsi_srv_mnt
1632 * we get this info from OSD just started */
1633 LASSERT(obd->obd_lu_dev);
1634 lsi->lsi_dt_dev = lu2dt_dev(obd->obd_lu_dev);
1635 LASSERT(lsi->lsi_dt_dev);
1637 dt_conf_get(NULL, lsi->lsi_dt_dev, &p);
1639 lsi->lsi_srv_mnt = p.ddp_mnt;
1645 /** Fill in the superblock info for a Lustre server.
1646 * Mount the device with the correct options.
1647 * Read the on-disk config file.
1648 * Start the services.
1650 int server_fill_super(struct super_block *sb)
1652 struct lustre_sb_info *lsi = s2lsi(sb);
1656 rc = lsi_prepare(lsi);
1660 /* Start low level OSD */
1661 rc = osd_start(lsi, sb->s_flags);
1663 CERROR("Unable to start osd on %s: %d\n",
1664 lsi->lsi_lmd->lmd_dev, rc);
1669 CDEBUG(D_MOUNT, "Found service %s on device %s\n",
1670 lsi->lsi_svname, lsi->lsi_lmd->lmd_dev);
1672 if (class_name2obd(lsi->lsi_svname)) {
1673 LCONSOLE_ERROR_MSG(0x161, "The target named %s is already "
1674 "running. Double-mount may have compromised"
1675 " the disk journal.\n",
1681 /* Start MGS before MGC */
1682 if (IS_MGS(lsi) && !(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOMGS)) {
1683 rc = server_start_mgs(sb);
1688 /* Start MGC before servers */
1689 rc = lustre_start_mgc(sb);
1693 /* Set up all obd devices for service */
1694 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC) &&
1695 (IS_OST(lsi) || IS_MDT(lsi))) {
1696 rc = server_start_targets(sb, lsi->lsi_srv_mnt);
1698 CERROR("Unable to start targets: %d\n", rc);
1701 /* FIXME overmount client here, or can we just start a
1702 * client log and client_fill_super on this sb? We
1703 * need to make sure server_put_super gets called too
1704 * - ll_put_super calls lustre_common_put_super; check
1705 * there for LSI_SERVER flag, call s_p_s if so.
1707 * Probably should start client from new thread so we
1708 * can return. Client will not finish until all
1709 * servers are connected. Note - MGS-only server does
1710 * NOT get a client, since there is no lustre fs
1711 * associated - the MGS is for all lustre fs's */
1714 rc = server_fill_super_common(sb);
1720 /* We jump here in case of failure while starting targets or MGS.
1721 * In this case we can't just put @mnt and have to do real cleanup
1722 * with stoping targets, etc. */
1723 server_put_super(sb);
1728 * Calculate timeout value for a target.
1730 void server_calc_timeout(struct lustre_sb_info *lsi, struct obd_device *obd)
1732 struct lustre_mount_data *lmd;
1736 bool has_ir = !!(lsi->lsi_flags & LDD_F_IR_CAPABLE);
1737 int min = OBD_RECOVERY_TIME_MIN;
1739 LASSERT(IS_SERVER(lsi));
1743 soft = lmd->lmd_recovery_time_soft;
1744 hard = lmd->lmd_recovery_time_hard;
1745 has_ir = has_ir && !(lmd->lmd_flags & LMD_FLG_NOIR);
1746 obd->obd_no_ir = !has_ir;
1750 soft = OBD_RECOVERY_TIME_SOFT;
1752 hard = OBD_RECOVERY_TIME_HARD;
1754 /* target may have ir_factor configured. */
1755 factor = OBD_IR_FACTOR_DEFAULT;
1756 if (obd->obd_recovery_ir_factor)
1757 factor = obd->obd_recovery_ir_factor;
1760 int new_soft = soft;
1761 int new_hard = hard;
1763 /* adjust timeout value by imperative recovery */
1765 new_soft = (soft * factor) / OBD_IR_FACTOR_MAX;
1766 new_hard = (hard * factor) / OBD_IR_FACTOR_MAX;
1768 /* make sure the timeout is not too short */
1769 new_soft = max(min, new_soft);
1770 new_hard = max(new_soft, new_hard);
1772 LCONSOLE_INFO("%s: Imperative Recovery enabled, recovery "
1773 "window shrunk from %d-%d down to %d-%d\n",
1774 obd->obd_name, soft, hard, new_soft, new_hard);
1781 obd->obd_recovery_timeout = max(obd->obd_recovery_timeout, soft);
1782 obd->obd_recovery_time_hard = hard;
1783 obd->obd_recovery_ir_factor = factor;
1785 EXPORT_SYMBOL(server_calc_timeout);